Train a Model

This tutorial will guide you through the process of training a model using DashAI.

Prerequisites

Before starting, make sure you have:

A dataset uploaded to DashAI
Understanding of your dataset’s columns (features and target)

Step-by-Step Guide

1. Access the Experiments Section

Click on the “Experiments” button in the navigation bar to access the experiments interface.

2. Create New Experiment

Click on the “New Experiment” button to start creating your experiment.

3. Select Task and Set Name

Write a name for your experiment
Select the appropriate task type for your problem:
- Text Classification
- Tabular Classification
- Image Classification
- Regression
- Translation
Click “Next” to proceed

4. Select Dataset

Choose the dataset you want to use for training
Click “Next” to proceed

5. Configure Input/Output and Splits

Select the input columns (features)
Select the output column (target)
Configure the train/test/validation splits
Click “Next” to proceed

6. Select and Configure Model

You can select multiple models to train and compare their performance. For each model:

Click on the “Select a model to add” button
Choose a model suitable for your task
Configure the model parameters
Repeat to add more models

Here are some recommended model combinations for each task type:

For Text Classification:
- DistilBERT + Bag of Words: Compare simple vs advanced approaches
- Multiple DistilBERT instances with different parameters
For Tabular Classification:
- Random Forest + Logistic Regression: Compare linear vs non-linear methods
- SVC + Random Forest: Compare different algorithmic approaches
For Image Classification:
- Different ViT configurations to compare performance
For Regression:
- Linear Regression + Random Forest: Compare simple vs complex relationships
- Multiple SVR instances with different kernels

Note

Training multiple models allows you to:

Compare performance across different architectures
Find the best model for your specific dataset
Understand trade-offs between complexity and performance
Experiment with different hyperparameter configurations

For a complete list of available models and their detailed documentation, see Models.

7. Model Optimization

DashAI provides automatic model optimization through:

Hyperparameter Optimization: Find the best parameters for your model automatically
- Each parameter combination is evaluated using the selected optimization metric
- Example parameters to optimize:
  
  SVC: Regularization parameter (C)
  
  Random Forest: Number of trees, maximum depth
Optimization Metrics: The metric used to compare different parameter combinations
- Classification:
  
  Accuracy: Overall correctness
  
  F1-score: Balance between precision and recall
  
  Precision: Exactness of positive predictions
  
  Recall: Completeness of positive predictions
- Regression:
  
  Root Mean Squared Error (RMSE): Root mean squared error
  
  Mean Absolute Error (MAE): Mean absolute error
- Translation:
  
  BLEU: Similarity between translations
  
  TER: Translation edit rate
Optimizers: Algorithms that guide the hyperparameter search
- OptunaOptimizer: Efficient parameter search using Bayesian optimization
- HyperOptOptimizer: Tree-structured Parzen Estimators for parameter search

Note

The optimization process will: 1. Try different parameter combinations 2. Evaluate each combination using the selected metric 3. Keep the parameters that achieve the best metric value

8. Train and review results

To train your models:

Click “Start” to start the training process
Wait for training completion
Review the model’s performance metrics
Compare with other runs

Tips and Best Practices

Start with a small dataset to test your configuration
Try different model architectures for your task
Use appropriate evaluation metrics for your problem
Save successful model configurations for future use
Consider using model optimization for better results

Troubleshooting

If training fails, check:
- Dataset format and preprocessing
- Model parameter values
- System resources (memory, GPU)
For poor performance:
- Review feature selection
- Adjust model hyperparameters
- Consider data preprocessing
- Try different model architectures